thebears/SearchInterface
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

YACWC

Browse Source
This commit is contained in:
Ishan S. Patel
2025-09-17 16:50:06 -04:00
parent 50376f71a8
commit 0fa6025514
8 changed files with 567 additions and 326 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
VectorService/util/embed_scores.py
View File

@@ -39,7 +39,6 @@ def get_matching_file_for_tstamp(target_tstamp, folder_scores):
web_name = 'media/'+os.path.basename(fname)
return dict(full_path = fname, path=web_name, timeoffset = offset)
def get_vec_rep_file_loc(c_dir):
vec_rep_file = os.path.join(c_dir, 'vec_rep.npz')
return vec_rep_file
@@ -92,7 +91,7 @@ def get_vector_representation(c_dir, force_compute = False, redis_key = 'compute
all_tstamps.append( [x.timestamp() for x in hh['frame_time']])
enu +=1
message = {'task':'VECTOR_CALC_IN_FOLDER_BUMP', 'progress': idx+1, 'how_many': len(sorted_videos), 'time': dt.datetime.now().timestamp()}
message = {'task':'VECTOR_CALC_IN_FOLDER_BUMP', 'when': c_dir, 'progress': idx+1, 'how_many': len(sorted_videos), 'time': dt.datetime.now().timestamp()}
r.rpush(redis_key, json.dumps(message))
if len(all_cat) == 0:
@@ -104,6 +103,7 @@ def get_vector_representation(c_dir, force_compute = False, redis_key = 'compute
np.savez(vec_rep_file, embeds = all_embeds, idces= all_idces, timestamps = all_times, source_files = all_source)
message = {'task':'VECTOR_CALC_IN_FOLDER_DONE', 'when': str(c_dir), 'time': dt.datetime.now().timestamp()}
r.rpush(redis_key, json.dumps(message))
return dict( embeds = all_embeds, idces= all_idces, timestamps = all_times, source_files = all_source)
@@ -113,7 +113,7 @@ def get_vector_representation(c_dir, force_compute = False, redis_key = 'compute
def get_scores_embedding_c_dir(c_dir, query_vector, redis_key = 'compute_log'):
vec_rep = get_vector_representation(c_dir, redis_key=redis_key)
query_scores = (query_vector @ vec_rep['embeds'].T).squeeze()
return query_scores
return vec_rep, query_scores
@functools.lru_cache
def get_query_vector(query):
@@ -226,8 +226,8 @@ def calculate_embedding_score_in_folder(og_dir, threshold, query = None, query_v
pass
vec_rep = get_vector_representation(c_dir, redis_key = redis_key)
query_scores = get_scores_embedding_c_dir(c_dir, tuple(query_vector.tolist()[0]), redis_key = redis_key)
# vec_rep = get_vector_representation(c_dir, redis_key = redis_key)
vec_rep, query_scores = get_scores_embedding_c_dir(c_dir, tuple(query_vector.tolist()[0]), redis_key = redis_key)
video_json_info = list()
idces_keep = np.where(query_scores > threshold)[0]

13
VectorService/vector_service.py
View File

@@ -1,3 +1,4 @@
from typing import Union, Optional, List
from pydantic import BaseModel
from fastapi import FastAPI, Request, Depends
@@ -24,9 +25,9 @@ r = redis.Redis(host='localhost', port=6379, db=15)
class VideosPostRequest(BaseModel):
query: str = "A cat and a human",
threshold: float = 0.10,
c_dirs: Optional[List[str]] = None,
query: str = "A cat and a human"
threshold: float = 0.10
c_dirs: Optional[List[str]] = None
task_id: str = 'compute_log'
@app.post("/videos.json")
@@ -61,9 +62,9 @@ async def videos_json(
]
print(','.join([str(x) for x in c_dirs]))
message = {'task':'SCHEDULED','when':[str(x) for x in c_dirs], 'time':time.time()}
r.rpush(task_id, json.dumps(message))
# print(','.join([str(x) for x in c_dirs]))
# message = {'task':'SCHEDULED','when':[str(x) for x in c_dirs], 'time':time.time()}
# r.rpush(task_id, json.dumps(message))?
for x in c_dirs: